BayesMix

BayesMix performs nonparametric Bayesian model-based inference to model gene intensity distributions and identify differentially expressed genes in microarray experiments.

Key Features:

• Nonparametric Bayesian mixture model: Employs a nonparametric Bayesian mixture of normal distributions to model the distribution of gene intensities without assuming a fixed number of components.
• Model-based inference: Performs fully model-based inference rather than empirical Bayes plug-in estimation for differential expression detection.
• Posterior simulation: Uses posterior simulation techniques analogous to those used in traditional nonparametric mixture-of-normal models.
• R and C implementation: Core algorithms are implemented as R functions with underlying C routines.
• Evaluation of posterior expected false discovery rates: Computes posterior expected false discovery rates to quantify error rates in findings.
• Inference without null samples: Enables inference when known null (non-differentially expressed) samples are unavailable.

Scientific Applications:

• Microarray differential expression: Identification of genes differentially expressed between normal and diseased tissues in microarray experiments.
• Colon cancer comparison: Application example includes comparing gene expression in normal tissue versus colon cancer samples.
• Error-rate assessment: Assessment of posterior expected false discovery rates in differential expression studies.
• Analysis without known nulls: Differential expression analysis in datasets lacking known non-differential (null) samples.

Methodology:

BayesMix models gene expression using a Bayesian framework with mixtures of normal distributions, performs posterior simulation for inference, is implemented via R functions and C routines, and has been validated through simulation studies.